Specific indefinites, presupposition, and scope* Bart Geurts

Department of Philosophy University of Nijmegen [email protected] www.phil.kun.nl/tfl/bart

Abstract In the fast few years it has been suggested by several independent sources that specificity and presupposition are kindred phenomena. My proposal is to relate specificity and presupposition to each other not by reducing the former to the latter, as in earlier accounts, but by subsuming them under a more comprehensive rubric, which I call ‘backgrounding’. I argue that this view doesn’t suffer from the problems other theories are faced with, and that it sheds new light on presupposition projection as well as on a number of phenomena that thus far lacked a systematic account.

*

Portions of this paper were presented at the 1999 ESSLLI workshop on focus and presupposition in multi-

speaker discourse (Utrecht) and the 2001 workshop on specificity and sentence type (Berlin); versions of the paper duly appeared in the proceedings of these workshops. The first version contained what in the meantime has appeared as a squib in Linguistic Inquiry (Geurts 2000a), where it is argued that specific indefinites cannot in general be construed in situ, or in other words, that movement of some sort is called for; here it will be taken for granted that this is so. For comments and discussion I am indebted to Reinhard Blutner, Paul Dekker, Brenda Kennelly, Rob van der Sandt, and Henk Zeevat.

The first version of this paper appeared in 1999 under the title ‘Specifics’ (in: B. Geurts, M. Krifka and R. van der Sandt, eds., Focus and Presupposition in Multi-Speaker Discourse, ESSLLI 99, Utrecht, pp. 99-129). This version, which is substantially contained in the first, was completed in 2000, has been in press ever since, and is still to appear in: R. Bäuerle, U. Reyle, and T.E. Zimmermann (eds.), Presuppositions and Discourse.

1 Introduction 1.1 Indefinites and scope The principal characteristic of specific indefinites is that they have a predilection for taking wide scope:1 (1 ) a. After all that effort and time they now don’t know where 40 per cent of it is. (New Scientist, 24 April 1999; the neuter pronoun refers to 182 kilograms of plutonium dumped into the Irish Sea by the Sellafield nuclear plant.) b. All critics who were invited to comment on some poems written by a 2-year-old bonobo hailed them as mature masterpieces. The indefinite NP ‘40 per cent of it’ in (1a) occurs within the syntactic scope of a negation sign and an attitude verb, but it is interpreted as if they weren’t there; for what the sentence means is something like: ‘40 per cent of the plutonium is such that they don’t know where it is.’ The same holds, mutatis mutandis, for the indefinite ‘a 2-year-old bonobo’ in (1b). Observations like these have been taken to show that specific indefinites always take widest scope, or even that they are referential expressions (e.g. Fodor and Sag 1982), but as examples given already by Kasher and Gabbay (1976) and Farkas (1981) demonstrate, neither claim is correct: (2 ) a. Now, after all that effort and time, they say they don’t know where 40 per cent of it is. (New Scientist, 24 April 1999)2 b. Each student has to come up with three arguments which show that some condition proposed by Chomsky is wrong. (Farkas 1981) c. The police report might indicate that Mary wants to marry a Swede. (Kasher and Gabbay 1976) The intended interpretation of (2a) presumably is that ‘they say that 40 per cent of the dumped plutonium is such that they don’t know where it is,’ and the same holds, mutatis mutandis, for the prepositional object in (2b) and a Swede in (2c). Hence, in each case the specific indefinite is interpreted as if it occurred midway between its actual surface position and the outermost scope-bearing expression. In all these examples there appears to be a mismatch between the position at which an indefinite appears and its preferred interpretation. Following many of the more recent contributions to the literature, I will assume that this is the hallmark of specificity (e.g. 1

I will argue eventually that specificity has nothing to do with scope, in the grammarians’ sense, but for the time

being I will use the notion as an expository device for distinguishing between readings. 2

(1a) occurred in a caption, and (2a) in the text, of the same article.

2

Abusch 1994, Reinhart 1997, Winter 1997, van Geenhoven 1998). Such mismatches are not the norm: indefinites are often interpreted in situ, and there is some reason for taking this to be the default option. The reason is that comparatively ‘neutral’, i.e. semantically attenuate, indefinites have a preference for in situ readings, as the following pairs illustrate: (3 ) a. Several students reported that they had been harassed by a professor. b. Several students reported that they had been harassed by a professor emeritus from the law faculty. (4 ) a. Several students reported that they had been harassed by professors. b. Several students reported that they had been harassed by professors wearing false beards and pink gowns. Both (3a) and (4a) are more likely to be understood with the sentence-final indefinite interpreted in situ. It is only when these expressions become ‘heavier’ that a specific reading is enforced, as (3b) and (4b) illustrate. Note, incidentally, that (4b) belies the popular view that bare plural indefinites are always construed in situ. It may be the case that they like such readings better than most other indefinites do, but bare plurals allow for specific construals, too. On the strength of these observations it may be assumed that in situ interpretations are the rule, and specific interpretations the exception. Van Geenhoven (1998) suggests, furthermore, that wide-scope construals of specific indefinites are preferred, ceteris paribus, to intermediate-scope construals. I believe that she right about this, though it must be conceded be that intuitions are rather delicate. At any rate, the argument must be along the same lines as previously: (5 ) a. Every city was represented by twelve athletes sponsored by a brewery. b. Every city was represented by twelve athletes sponsored by a local brewery. (6 ) a. Every newspaper featured multiple reviews of a gothic novel. b. Every newspaper featured multiple reviews of a gothic novel written by its editorin-chief. Setting in situ readings aside, it seems to me that in the (a) sentences there is a preference for construing the sentence-final indefinites as having wide, rather than intermediate, scope. The balance tips, however, when the indefinites are enhanced with material enabling a link with the universally quantified subject, as the (b) examples demonstrate. These observations support van Geenhoven’s claim that, all else being equal, wide-scope readings are more easily obtainable than intermediate-scope ones. It bears emphasizing that these preferences hold

3

ceteris paribus only, and are easily overridden by considerations of plausibility, as indeed the examples in (3) to (6) demonstrate. We thus arrive at the following preference order on the range of possible interpretations of indefinite NPs: in situ < wide scope < intermediate scope In the second half of this paper I endeavour to show how this pattern can be accounted for. 1.2 No need to know There is a widespread opinion that in order for an indefinite NP to be used with a specific interpretation, the speaker must have a particular individual in mind (e.g. Kasher and Gabbay 1976, Fodor and Sag 1982, Manga 1996, Kratzer 1998, Yeom 1998, van Rooy 1999). It might be thought that this explains the unmistakable family resemblance between specific indefinite NPs, on the one hand, and definite NPs, on the other (which will be documented at some length in the next section). Just as a speaker employs the definite article to signal that an individual is given as part of the common ground between him and the hearer, he employs a specific indefinite if he wants to indicate that an individual is known to him, though not to his audience. In short: while definiteness implies givenness to speaker and hearer, specificity implies accessibility to the speaker alone. (For obvious reasons, there are no linguistic devices for signalling that an individual is accessible to the hearer alone.) This view on specificity is untenable. As Haspelmath (1997) points out, there are many languages that allow indefinite NPs to be morphologically flagged as ‘unknown to the speaker’, but the use of such flags doesn’t entail non-specificity. For example, German ‘irgendein N’ conveys that the speaker doesn’t know the N in question, but may well be used specifically: (7 ) Wilma hat vor, irgendeinen Schweden zu heiraten. Wilma intends some-or-other Swede to marry. Even in the absence of explicit morphological clues, there are many cases in which it is simply false, intuitively speaking, that the witness of a specific indefinite must be known to the speaker. This is especially problematic when specific indefinites take intermediate scope, but these are not the only cases. Consider (1a), for example. It would be patently wrong to say that the author of this sentence must have had a particular portion of plutonium in mind; yet there can hardly be any doubt that the indefinite ‘40 per cent of it’ is being employed in a specific sense. Whatever it may be, having something in mind is not a prerequisite for specificity.

4

Having arrived at this conclusion, we should ask ourselves how we can recognize specificity in the absence of telltale scope-bearing expressions. The answer to this question, I submit, is that by and large we can’t. That is to say, the chief problem for a theory of specificity is to account for the interaction between specific indefinites and further scopebearing expressions occurring in the same sentence. Apart from that, there is little in the way of systematic evidence for a theory of specificity to build on, though the following type of data might be an exception: (8 ) At the party, Fred danced with an Irish woman, and so did Barney. This sentence may or may not be construed as implying that Fred and Barney danced with the same woman, and if this is to do with the fact that the indefinite ‘an Irish woman’ is either specific or non-specific, as suggested by Kasher and Gabbay (1976), then this is a case in which specificity manifests itself even in the absence of other scope-bearing expressions. Note, however, that in cases like (8), one purported reading is always entailed by the other, which makes it hard to prove that these sentences are truly ambiguous. 1.3 Similarities between specific indefinites and definites It was hinted already that, in certain respects, there is a resemblance between specific indefinites and definite expressions. In fact, the similarities are quite striking, as the following observations will demonstrate, and if these facts may be taken at face value, any theory of specificity worth its salt should be able to explain why definites and specific indefinites are so much alike. 1.3.1 Scope The hallmark of specific indefinites is that they tend to take scope over anything else in the sentence, which is characteristic of definites, too. One example will suffice to illustrate this well-worn observation: (9 ) All critics who were invited to comment on some poems written by the Danish chancellor hailed them as mature masterpieces. (cf. (1b)) This is most likely to be read as implying that Denmark has a chancellor who wrote all the poems presented to the various critics. Of course, definites can take ‘intermediate scope’, too, as (10) demonstrates: (10 ) All critics who were invited to comment on some poems written by their spouses hailed them as mature masterpieces.

5

If the possessive pronoun is bound by the subject NP, it is of course impossible to obtain a wide-scope reading for the definite expression their spouses; but an intermediate reading remains feasible — indeed, it is the most natural reading in this case. One respect in which definites differ from indefinites at large is that it is quite difficult to obtain something akin to in situ readings for the former, whereas we have seen that the latter prefer such readings. Narrow-scope readings for definite NPs do occur, though: (11 ) That wasn’t Fred’s wife, you blockhead: Fred isn’t even married! But such examples are clearly marked (see Geurts 1998a for further discussion). Hence, although definites and indefinites are quite similar in the way they interact with scope-bearing expressions, their preferences in this regard are different. To summarize: definites: indefinites:

wide scope < intermediate scope < in situ in situ < wide scope < intermediate scope

1.3.2 Partitives As Ladusaw (1982) was the first to point out, the nominal constituent of a partitive PP must be definite or specific; non-specific indefinites and quantified NPs are not allowed in this position: (12 ) Fred is one of {the / several / *most / *all / *sm / *∅} employees who will be fired. Here sm represents unstressed some, which has a distinct preference for a non-specific reading, like the bare plural, indicated by ‘∅’. 1.3.3 Indefinite this Although formally this is a definite article, it sometimes appears to function as if it were indefinite (see Prince 1981 for discussion): (13 ) There is this giant spider in the cupboard. When used in this manner, this-NPs function as indefinites because, intuitively, they introduce discourse entities that are new, an intuition which is corroborated by the following example:

6

(14 ) ?Yesterday, our little daughter brought [a giant spider]i into the house, and now there is [this giant spider]i in the cupboard. It is not possible, apparently, for the this-NP to refer back to the spider introduced in the first conjunct, which confirms the impression that ‘this giant spider’ functions as an existential expression. Furthermore, indefinite this-NPs behave more like specific than non-specific indefinites, because they typically take wide scope: (15 ) a. If this giant spider is in the cupboard, Betty will go berserk. b. There is a giant spider at large, and if it is still in the cupboard, Betty will go berserk. c. If there is a giant spider in the cupboard, Betty will go berserk. (15a) is more or less synonymous with (15b), rather than (15c), which is precisely what one should expect if ‘this giant spider’ were specific. These observations indicate that indefinite this-NPs are expressions that are marked for definiteness but function as specific indefinites. It is hard to see how this ensemble of form and function could occur unless specificity and definiteness are kindred phenomena. 1.3.4 Cross-linguistic evidence Perhaps the most telling evidence is that in language after language definiteness and specificity are lumped together into the same morpho-syntactic rubric. I will give a handful of more or less arbitrarily chosen examples. • Bemba: In Bemba, a Bantu language, there is a class of nominal prefixes of the form consonantvowel, and another class of the form vowel-consonant-vowel. The former are used to mark non-specific indefinites, while the latter alternatively convey definiteness or specificity. The following examples are from Givón (1978); here and in the following glosses are reproduced from the original source: (16 ) a. Umu-ana a-a-fwaaya ci-tabo. vcv-child he-past-want cv-book ‘The child wanted a book (be it any).’ b. Umu-ana t-a-à-somene ci-tabo. vcv-child neg-he-past-read cv-book ‘The child didn’t read a/any book.’

7

c. Umu-ana a-a-fwaaya ici-tabo. vcv-child he-past-want vcv-book ‘The child wanted the book’ or ‘The child wanted a specific book.’ • Samoan: Samoan is similar to Bemba in that it has two articles, one of which signals non-specific indefiniteness, while the other combines specificity and definiteness (examples from Lyons 1999): (17 ) a. Sa i ai le uluga¯li‘i‘o Papa le tane a ‘o Eleele le fafine. Past exist Art couple Pres P. Art husband but Pres E. Art woman ‘There was a couple, Papa, the husband, and Eleele, the wife.’ b. ‘Au-mai se niu. take-Dir Art coconut ‘Bring me a coconut.’ • West Greenlandic Inuit: In West-Greenlandic Inuit, an ergative language, transitive verbs may become intransitive by incorporating their objects. This shows itself, among other things, in the case marking on the subject, which is absolutive for intransitive, and ergative for transitive subjects. Moreover, it is only in transitive constructions that verbs bear object-agreement markers. The object of a transitive construction receives absolutive case, and may be either specific or definite, while incorporated objects are non-specific. According to Manga (1996), this is typical of ergative languages. The following sample of West-Greenlandic Inuit is from van Geenhoven (1998): (18 ) a. Angunguaq tikip-p-u-q. A.Abs arrive-Ind-Intr-3sg ‘Angunguaq arrived.’ b. Angunguaq aalisakka-mik neri-v-u-q. A.Abs fish-Inst.sg eat-Ind-Intr-3sg ‘Angunguaq ate fish.’ c. Arnajaraq aalisaga-si.nngi-l-a-q. A.Abs fish-buy-Neg-Ind-Intr-3sg ‘It is not the case that Arnajaraq bought {a / more than one} fish.’ d. Angunguu aalisagaq neri-v-a-a. A.Erg fish-Abs eat-Ind-Tr-3sg.3sg ‘Angunguaq ate the/a particular fish.’

8

• St’át’imcets: St’át’imcets (Lillooet Salish) features an indefinite article which can only occur within the scope of a negative expression, a question, a modal, and so on. In the absence of such operators another article must be used, which has a specific-definite function. The following examples are from Matthewson (1999): (19 ) a. Cw7aoz kw-s áts’x-en-as ku sqaycw. Neg Det-Nom see-Tr-3Erg Det man ‘S/he didn’t see any men.’ b. *Áts’x-en-as ku sqaycw. see-Tr-3Erg Det man ‘S/he saw a man.’ (20 ) a. Húy-lhkan ptakwlh, ptákwlh-min lts7a ti smém’lhats-a … going.to-1sg.Subj tell.story tell.story-Appl here Det woman.Dimin-Det ‘I am going to tell a legend, a legend about a girl …’ b. Wa7 ku7 ílal láti7 ti smém’lhats-a Prog Quot cry Deic Det woman.Dimin-Det ‘The girli was crying there.’ This sample will suffice to show that many languages treat definiteness and specificity as related notions, which together stand in opposition to non-specific indefiniteness. In conjunction with the evidence of the preceding sections, these data raise the question what it is that definites and specific indefinites have in common. In my opinion, one of the main criteria for assessing theories of specificity should be how good their answers to this question are. 1.4 Specificity and distributivity It has been argued by Reinhart (1997) and Winter (1997) that specific indefinites which have escaped from a scope island don’t allow for a distributive interpretation. Reinhart credits Ruys with this insight; Winter attributes it (collectively) to Ruys and himself. Reinhart cites example (21a) from a manuscript by Ruys: (21 ) a. If three relatives of mine die, I will inherit a house. b. There are three relatives of mine such that, if they all die, I will inherit a house. c. There are three relatives of mine such that, if any of them dies, I will inherit a house. On the most likely reading of (21a) the indefinite ‘three relatives of mine’ is construed with narrow scope, but if it gets a specific reading and outscopes the if-clause, then according to

9

Reinhart, Ruys, and Winter, it can only be understood collectively. That is to say, if the indefinite is specific, (21a) is synonymous with (21b), not (21c). This observation is not quite correct, however; what Reinhart et al. have found is not a lawful correlation but merely a trend. First, as noted by Matthewson (1999), there are native speakers of English who manage to obtain a distributive reading for (21a), and the same holds for parallel sentences of other languages. Secondly, van Geenhoven (1998) points out that intuitions shift markedly when we vary the example. Thus it appears to be easier to get a distributive reading for the following sentence: (22 ) If some relatives of mine invite me for dinner, I will panic. In short, although in environments like (21a) or (22) specific indefinites seem to prefer collective construals, specificity does not entail collectivity. This is bad news for two rather different theories of specificity. On the one hand, theories that seek to deal with specificity with the help of quantifier raising will be embarrassed by the fact that specific indefinites disprefer non-distributive readings. On the other hand, theories that rely on choice-functions instead of quantifier raising will find it quite difficult to explain the distributive readings — a point which Winter (1997) emphasizes, because he is confident, apparently, that such readings don’t occur (for further discussion, see Geurts 2000a). There is one family of theories that can account for distributive as well non-distributive readings: these are theories which, on the one hand, resemble the quantifier-raising approach in that their account of specificity is based on movement, while, on the other hand, they agree with the choice-function approach that indefinites aren’t quantifier expressions. Two such theories are discussed below.

2 The binding theory of presupposition In the remainder of this paper I present a unified account of specificity and presupposition, which is based upon the binding theory of presupposition, so before we move on I want to quickly recapitulate the main tenets of that theory; for more extensive discussion, see van der Sandt (1992), Geurts (1999), and Geurts and van der Sandt (1999). The binding theory is an extension of discourse representation theory (Kamp 1981), and consists of three principal claims. The first of these is that anaphora is a species of presupposition, and that the standard presupposition-inducing expressions (such as definite NPs, factives, transition verbs, and so on) differ from pronominal anaphors mainly in that they possess a richer semantic content. This difference explains why in general presupposition inducers, unlike anaphoric pronouns, can be interpreted by way of accommodation, which is the second key notion in the theory. Finally, it is assumed that the process of presupposition projection is subject to certain constraints. It is the status of these constraints that will be especially important in the following.

10

Formulated in procedural terms, the binding theory predicts that if an utterance contains a presupposition-inducing element, the hearer will initially attempt to bind the presupposition to a suitable antecedent, just as he would try to bind an ordinary anaphor. If the presupposition cannot be so bound, it will be accommodated, i.e. it will be inserted in some accessible discourse representation structure (DRS). In general the number of positions at which a presupposition may be accommodated is greater than one, and if it is the choice is restricted by various constraints, but before I turn to these, let me first illustrate the workings of the theory: (23 ) If Fred is gay, then his son is gay, too. This sentence contains (at least) two presupposition-inducing expressions: the definite NP his son, which triggers the presupposition that Fred has a son, and the focus particle too, which triggers the presupposition that someone other than Fred’s son is gay. Note that the first presupposition is ‘inherited’ by the sentence as a whole, while the second one is not: normally speaking, an utterance of (23) would license the inference that (according to the speaker) Fred has a son, but not that someone else besides Fred’s son is gay. The binding theory accounts for these observations as follows. Suppose that the grammar assigns (23) the intermediate semantic representation in (24a). I assume for convenience that most interpretative problems have been cleared out of the way already, and that the only thing that remains to be done is resolve the presuppositions triggered by his son and too, which are marked out by single and double underscores, respectively. 3 (24 ) a. b. c. d.

[x: Fred(x), [: gay(x)] ⇒ [u, v: x’s-son(u), gay(v), v ≠ u, gay(u)]] [x, u: Fred(x), x’s-son(u), [: gay(x)] ⇒ [v: gay(v), v ≠ u, gay(u)]] [x, u: Fred(x), x’s-son(u), [v: v = x, gay(x), gay(v), v ≠ u] ⇒ [: gay(u)]] [x, u: Fred(x), x’s-son(u), [: gay(x)] ⇒ [: gay(u)]]

(24a) is the initial semantic representation correlated with (23), in which two presuppositions remain to be resolved. One of these, that Fred has a son, cannot be bound, and therefore must be interpreted by way of accommodation. Now there is a general constraint on presupposition projection to the effect that any presupposition prefers to be projected to as high a position as possible, and accordingly our first presupposition is accommodated in the principal DRS, which yields (24b). The remaining presupposition, triggered by the focus particle, can be bound in the antecedent of the conditional; this results in (24c) which, assuming that Fred and his son are different persons, is equivalent to (24d).

3

Officially, I maintain that the presupposition triggered by too is not quite as simple as what is assumed in the

following analysis. See van der Sandt and Geurts (2001) for discussion.

11

The binding theory may be summed up in the following three principles: (A) (B) (C)

Presuppositions must be projected (i.e., bound or accommodated). Binding is preferred to accommodation. A presupposition must be projected to the highest possible DRS.

It will be evident that none of these principles is absolute, although the first two may be more absolute than the last one. They are all subject to general constraints on interpretation, which require that an interpretation be consistent, coherent, and so on. Before these principles come into play, presuppositions are merely representational structures, and are therefore completely inert. Principle A drives away the inertia by insisting that presuppositions be either bound or accommodated. Principle B captures the insight that accommodation is a repair strategy: in principle, a presupposition wants to be bound, but if it cannot be bound it will be accommodated. Principle C may be viewed as a generalization of a constraint first proposed by Heim (1983). Heim distinguishes between two types of accommodation: global and local. In terms of the present framework, a presupposition is accommodated globally if it goes to the principal DRS, and locally if it is accommodated in the DRS where it was triggered. Heim’s proposal is that, in general, global accommodation is preferred to local accommodation, and principle C generalizes this in two ways. First, this principle applies not only to accommodation but to projection in general. This makes some difference from an observational point of view (though not much), and it is surely more attractive conceptually speaking. Secondly, although it is possible to capture Heim’s distinction between global and local accommodation in our framework, the distinction as such doesn’t play a role in the theory. In general, there is a line of accessible DRSs in which a presupposition can be accommodated, the two ends of this chain being the main DRS and the DRS where the presupposition arises. Global and local accommodation are just convenient labels for referring to accommodation in these DRSs, but they do not denote special processes. I should like to stress that the fundamental insight underlying this treatment of presuppositions is not a controversial one. It is that presupposed information is information that is presented as given. Most extant theories of presupposition accept this premise, too. What distinguishes the binding theory from other accounts is just that it doesn’t draw a sharp line between presupposition and anaphora. Hence, although the choice of framework is essential in some respects, the gist of my analysis of specific indefinites could be expressed in other frameworks, too. To say that presupposed information is presented as given is not to say that it is given. Indeed, the concept of accommodation merely puts a label on the observation that speakers are wont to exploit (in Grice’s sense) presupposition-inducing expressions in order to convey information that is new. The point is a familiar one, I take it, but it deserves to be stressed, because it is sometimes thought that accommodation will be the weak spot of any theory of presupposition that adopts the notion, as most of them do (see Abbott 2000 for a recent attack

12

along these lines). Even if it could be demonstrated that, say, definite NPs are regularly used to refer to entities that are new (and Abbott maintains that this has been demonstrated), that wouldn’t even begin to show that the standard view of presupposition is on the wrong track. It would merely corroborate what we knew already, namely that speakers are adept at exploiting (still in the Gricean sense) linguistic devices for their purposes.

3 Accommodating indefinites In the past few years, it has been suggested by several independent sources that specificity should be handled in terms of, or at least in conjunction with, presupposition projection (Cresti 1995, Yeom 1998, van Geenhoven 1998). This is an attractive idea, as I will try to show, but it requires a dramatic change of perspective, too, because it implies that specificity is an essentially pragmatic phenomenon. Following these developments, I will present my own unified theory of presupposition and specificity in the next section. In many respects, my account is related as well indebted to van Geenhoven’s, which will therefore be discussed first. 3.1 Incorporation vs. accommodation The majority position in the literature on specificity is that indefinites are ambiguous between specific and non-specific readings. Van Geenhoven (1998) doesn’t take exception to this view, but she develops it in an entirely new way. According to van Geenhoven, non-specific indefinites are ordinary predicates, which neither possess quantificational force nor introduce reference markers or anything of the sort. If the indefinite in (25a), for example, is interpreted non-specifically, it doesn’t have narrow scope; indeed, it doesn’t have scope at all because it is semantically incorporated by the verb, as suggested by the paraphrase in (25b): (25 ) a. Every man loves a woman. b. Every man is a-woman-lover. If, on the other hand, an indefinite gets a specific reading, its semantic representation is rather different. Specific indefinites are analysed in accordance with the standard DRT doctrine on indefinites, save for the fact that it is stipulated that they must be accommodated. Or in other words, if a woman in (25a) is specific, it is treated as if it were a presupposition-inducing expression whose presupposition has the peculiarity that it doesn’t want to be bound. Hence, the indefinite is dealt with in two steps. First, the grammar produces the initial discourse representation in (26a), in which the semantic correlate of a woman is marked as specific, and then this representation is fed into the projection mechanism of the binding theory, which treats the indefinite description as it would treat any (other) presuppositional expression,

13

except that it cannot be bound. Consequently, it must be accommodated, and since there is a general preference for accommodating things at the highest level of representation, it is predicted that the resulting interpretation will be (26b). (26 ) a. [: [x: man(x)]〈every x〉[u: woman(u), x loves u]] b. [u: woman(u), [x: man(x)]〈every x〉[: x loves u]] I find this analysis appealing for a number of reasons. To begin with, it comes essentially for free, because all the machinery it employs is already in place, as it is required anyway for dealing with presupposition projection. Secondly, van Geenhoven’s proposal explains the parallels as well as the differences between definites and specific indefinites. The reason why definites and specific indefinites are so similar is that they are interpreted by the same projection mechanism; the main difference is that definites want, and specific indefinites don’t want, to be bound. Thirdly, the theory accounts in a principled way for the puzzling pattern of interpretations discussed in section 1.1, which I repeat here for ease of reference: in situ < wide scope < intermediate scope According to van Geenhoven, indefinites are ambiguous between a specific and a nonspecific reading, and if it may be assumed that the latter prevails by default, then an in situ construal is preferred to a reading that involves movement, and if an indefinite gets a specific reading, principle C of the binding theory entails a preference for a wide-scope as opposed to an intermediate-scope reading. Although van Geenhoven’s theory hinges on the premise that specific indefinites are construed by way of movement, it should be stressed that this account has nothing to do with quantifier raising (or, for that matter, any other of the standard techniques for dealing with quantifier scope). Presupposition projection is a pragmatic affair, and therefore van Geenhoven’s proposal can only be seen as an attempt at dealing with specificity in pragmatic terms. Quantifier raising, in contrast, takes place at or near the syntax-semantics boundary, so a theory based on raising implies that specificity is a grammatical phenomenon. 3.2 Objections Although I applaud van Geenhoven’s pragmatic turn, and agree with the fundamental intuition underlying her theory, I have two objections, one of which I consider to be particularly serious. To begin with the major problem, I maintain that van Geenhoven’s analysis is conceptually incoherent. To my mind, the very idea of a class of expressions that insist on being interpreted by way of accommodation is a contradiction in terms. Accommodation is a repair strategy by definition. A speaker who presupposes that ϕ presents ϕ as given, and if it is not given it is at the hearer’s discretion whether or not he wants to play

14

along by accommodating ϕ. Therefore, accommodation isn’t anything like an ordinary rule of interpretation; it is a fall-back option, and if one wants to postulate a linguistic category that selects for this option, there is a fair amount of explaining to do. But can’t we simply broaden the concept of accommodation by ruling that it applies not only to presuppositions but to certain other types of information, as well? We can, of course, but there is a price to pay. A broadening of the notion of accommodation entails that we forfeit a powerful explanatory lever in our theory of presupposition projection. For we then will have to come up with new answers to such questions as: What justifies accommodation?, Why is binding preferred to accommodation?, and so on. And as long as I don’t see how these questions might be answered, I am not willing to pay this price. My slightly less urgent complaint concerns van Geenhoven’s assumption that indefinites are systematically ambiguous between specific and non-specific readings. Notwithstanding the fact that this assumption is commonplace in the literature, I don’t believe there is much independent evidence to support it, but that is as it may be, because nobody would deny that ambiguities are ugly and should be avoided at practically any cost. And, come to think of it, one should expect that a specific/non-specific ambiguity can be avoided in a framework based on the insight that specificity is a pragmatic phenomenon.

4 Specificity and backgrounding My proposal is to relate specificity and presupposition to each other, not by reducing the former to the latter, as van Geenhoven has tried to do, but by subsuming them under a more comprehensive rubric, which I call ‘backgrounding’. I will argue that this view doesn’t suffer from the shortcomings discussed in the foregoing, and, furthermore, that it throws a new and perhaps brighter light on presupposition as well as on a number of phenomena that thus far lacked a systematic account. 4.1 Foreground and background Following Foley and van Valin (1985) and Foley (1994), among others, I understand the opposition between foreground and background distinction purely in terms of informational prominence, where prominence is a relational rather than an absolute notion. By uttering a sentence a speaker typically conveys a considerable amount of information, only a small portion of which is central to his concerns. The remainder is backgrounded information: ancillary matter that merely serves to anchor the foregrounded information to the context, or information which is brought in en passant. Backgrounded information is not necessarily unimportant, but it is of secondary interest in relation to foregrounded information. Thus the notion of background is primarily a negative one: backgrounded information is what remains when foregrounded information is taken away. It may well be, therefore, that it is impossible

15

to provide a single positive description covering all sorts of background information. But no matter how many reasons for, or ways of, backgrounding there may be, I will suggest that at least some interpretative mechanisms do not discriminate between them. A further, and crucial, negative characteristic of my notion of background is that it doesn’t entail givenness; only the converse is true. Backgrounded information may be given, or presented as given, but new information is not necessarily foregrounded. For example, enclosing new information in (intonational or orthographic) parentheses often serves to indicate that it is of secondary importance, which is to say that it is backgrounded, not that it is presented as given. My notion of background is clearly related to Abbott’s (2000) ‘nonassertion’ and Horn’s (2000) ‘assertoric inertia’. The basic intuition in each case is that the main point of an utterance enjoys a special pragmatic status, while the remainder is, in some sense, downgraded. What distinguishes my concept from the other two is mainly that its interpretative effects are more explicit (see below). Apart from that I prefer to avoid the notion of assertion in this connection, because otherwise I would have to assume, contrary to what I take to be linguistic common sense, that assertions may occur in syntactically embedded positions. Although the distinction between foreground and background may be signalled by intonational means, I don’t want to make any substantial claims about the relationship between intonation and foreground/background. However, I should like to note that the correlation between intonational prominence and foregrounding is imperfect, at best. This observation is not new, but I feel it bears emphasizing nonetheless. Consider the following example: (27 ) The course on postmodern theology will be given by [the dean]F. Suppose, for enhanced clarity, that this is an answer to the question ‘Who is teaching the course on postmodern theology this year?’, so we can be sure that the non-focused part of (27) is given, and therefore backgrounded. Now of course the focused part is (presented as) given, too, simply by virtue of the fact that it is a definite NP. But surely everything in this statement cannot be given? The solution to this puzzle is not so hard to find: the focus on the dean doesn’t highlight the dean, but rather the fact that it is he who will be teaching the course on postmodern theology. The dean is given; that he will play a certain role is foregrounded. If backgrounded information need not be given, there is no reason why it couldn’t be marked as new. I want to suggest that this is not just an abstract possibility: it does happen that backgrounded information is marked as new; for this is precisely what specificity comes down to. Intuitively, specific indefinites carry information that is used to set the stage for the utterance’s main point. Like presuppositions, specific indefinites are separate units of

16

communication, to be integrated into the context before the remainder of the utterance is processed; this is why semantically incorporated indefinites cannot be specific.4 4.2 Accessibility and the Buoyancy Principle An utterance is always interpreted within a context, and broadly speaking utterances and contexts interact with each other in two ways: the context affects the interpretation of an utterance, which in its turn changes the context in which it occurs. In DRT the context of utterance is pictured as a line of accessible DRSs, and therefore the notion of accessibility is of central importance to DRT (as it is, mutatis mutandis, to all dynamic theories of meaning). What, exactly, is accessibility? From a technical point of view this question is not so hard to answer, but when we interpret the question as being about the theoretical status of the accessibility relation, many different answers are possible. In Kamp’s (1981) original version of DRT, accessibility was associated with anaphoricity in the sense that it was only used for constraining the interpretation of anaphora: an anaphoric pronoun had to find its antecedent in an accessible DRS. In later versions of the theory, the notion of accessibility gradually assumed a much broader significance. Thus, as we have seen in section 2, in the binding theory of presupposition accessibility demarcates what is given at the point where an expression occurs. I believe that an even broader view is called for, and that the accessible domain must be seen as the background against which an utterance is interpreted, where ‘background’ is to be understood as explained above. When we thus broaden our perspective on the significance of accessibility, it is only to be expected that some of the principles of interpretation hitherto cast in terms of accessibility will have to be generalized. This applies, in particular, to principle C of the binding theory, which I propose to supplant with the following: The Buoyancy Principle Backgrounded material tends to float up towards the main DRS. Strictly speaking, the Buoyancy Principle isn’t part of our theory of presupposition projection, because it is not specifically about presuppositions, so all that remains of the original binding theory is two ‘axioms’, one saying that presuppositions want to be bound, the other, that presuppositions that cannot be bound may be accommodated. The theory’s predictions aren’t affected by this change, although they are now seen in a somewhat different light. In particular, I am no longer committed to the claim that presuppositions tend to take 4

I suspect, moreover, that any indefinite that is not semantically incorporated is eo ipso specific. This position

entails, however, that generic indefinites are generally specific, and thus runs up against the currently prevailing view, as expressed, for example, by Krifka et al. (1995), that indefinite generics in the singular are never specific. I believe this view can be disproved, but will not try to do so here.

17

‘wide scope’ because they are presuppositions; it is rather because they are backgrounded, and therefore subject to the Buoyancy Principle, that they gravitate towards the principal DRS. But as far as the theory of presupposition is concerned, the proposed modification isn’t exactly a volte-face. Still, this relatively minor amendment may turn out to be more consequential than one should think, because it invites a rethinking of the binding theory’s treatment of at least some presupposition triggers, as I will argue in section 5. The concept of buoyancy itself is discussed at greater length in Geurts (2000b). 4.3 Explaining specificity In keeping with DRT orthodoxy, I regard indefinites as property-denoting expressions that receive existential import when they occur in argument positions. The main advantage of this division of labour is that it makes for a uniform analysis of indefinites occurring in argument positions and indefinite non-arguments, such as predicate nominals, for example. To illustrate, it allows us to maintain that a ventriloquist has the same meaning in both of the following sentences: (28 ) a. Barney is a ventriloquist. b. Betty is married to a ventriloquist. In (28a) as well as in (28b), a ventriloquist merely denotes a property, but only in the latter case is this property applied to a reference marker introduced by the verb. I will assume that, if this happens, the reference marker in question is labeled as new. There are various ways of accounting for this feature (if it is one), but that is a topic I don’t want to go into here. Unlike Reinhart, van Geenhoven, and many others, I deny that indefinites are ambiguous between a specific and a non-specific reading: indefinites always denote properties. If an indefinite occurs as an argument it may be construed as specific or non-specific depending on whether is backgrounded or not, which is to say that the choice is a pragmatic one. Of course, to say that a given aspect of interpretation is a pragmatic one is not to deny the possibility that it is conventionally marked in some languages. In this respect, specificity is in the same boat as definiteness, which is a pragmatic notion, too, and is conventionally marked in some, though by no means all, languages. Following the general consensus, I take it that by default indefinites are construed nonspecifically, and the most natural way of accounting for this is by assuming that, all things being equal, an indefinite will tend to be construed as part of the foreground because it carries new information. I still deny, of course, that new information is always foregrounded, but it is only natural that the former status tends to be escorted by the latter. It is only under special circumstances that new information is backgrounded, and if this happens, the expression in question is specific.

18

We are now all set to explain the main facts about specificity, beginning with the interaction between indefinites and (other) scope-bearing expressions. We have just seen why indefinites prefer to be construed non-specifically; this is, I suggested, because they tend to be part of the foreground. But if they are backgrounded, the Buoyancy Principle applies, which is to say that, other things being equal, they will take wide scope, and only if all things aren’t equal will they take intermediate scope. This is precisely the order of preferences that we wanted to account for. Secondly, the similarities between definites and specific indefinites fall into place, too, because both types of expressions convey backgrounded information. Thirdly, and by the same token, it is only to be expected that there will be languages which lump together specificity with definiteness, assigning the two functions a single article or case marker, say. On the present account, such conventional devices receive a straightforward interpretation: they signal that something is part of the background. Thus a vcv-prefix in Bemba, for example, isn’t ambiguous in any way; it just serves to indicate that the expression it attaches to is backgrounded. The partitive constraint is explained along the same lines. It is reasonable to suppose that, in an expression of the form ‘Det α of β’, the main duty of β is to help identify the intended α, and is therefore backgrounded (cf. e.g. Kuno 1987). So, properly understood, the partitive constraint is not that β must be either definite or specific, but rather that it must be backgrounded. This explains why definites and specific indefinites can occur in partitive constructions, while quantifiers and non-specific indefinites cannot. 4.4 Definites and indefinites in attitude contexts Consider the following sentence: (29 ) Barney believes that the university hangman is after him. On the de dicto reading of the definite NP, this statement entails that Barney believes that the university employs a hangman, who is after him. Thus construed, the sentence does not imply that there is a university hangman, as it does on the de re reading, which says of the university hangman that Barney believes that he is after him. Now compare (29) with the following minimal variant: (30 ) Barney suspects that the university hangman is after him. Here we have the familiar de re and de dicto readings, as in (29), but now we discern a third type of reading, as well, which can be paraphrased like this: (31 ) Barney believes that the university employs a hangmani and he suspects that hei is after him (where a hangman is to be read de dicto).

19

Intuitively speaking, this reading is de re and de dicto at the same time. It is de re in that, some way or other, the university hangman is not interpreted as dependent on suspects, though it is confined to Barney’s beliefs — which is remarkable of course, because there is no explicit reference to Barney’s beliefs in (30). These observations about definite NPs extend not only to other presupposition-inducing expressions, but to indefinite NPs, as well: (32 ) Barney suspects that an evil spirit is haunting him. Here, too, we have three readings, one which is de re, another which is de dicto, and a mixed reading which can be paraphrased thus: (33 ) Barney believes that there is an evil spiriti and he suspects that iti is after him (where an evil spirit is to be read de dicto). In Geurts (1998a, 1999) I have ventured to give an in-depth analysis of the interaction between attitude verbs and presupposition-inducing expressions, and as that account carries over to specific indefinites without further ado, I will only give a quick outline here. The main idea is that the meanings of attitude verbs such as suspect, hope, want, and so on, are defined in terms of their subjects’ doxastic states (cf. Stalnaker 1984, Heim 1992). For example, to say that a suspects that ϕ is to say, approximately, that among the worlds compatible with what a believes, those worlds which a considers to be more likely all make ϕ come out true. More generally, doxastic states provide the context against which we attribute not only suspicions but also hopes, desires, and so forth. In DRT terms this means, for (30) and (32), that a representation of Barney’s doxastic state acts as the intermediate context between the main context and the suspicion that is being attributed to him. Therefore, a presupposition triggered within the syntactic scope of suspect will project to the main context by default, or if that option is barred for some reason or other, it will go to the DRS representing Barney’s doxastic state. Similarly, a specific indefinite will either be construed in the main context or as part of Barney’s beliefs. Hence, the various readings available for (30) and (32) are predicted in a uniform fashion. I have shown elsewhere that attitude reports are a crucible for theories that always want to construe indefinites in situ; such theories, I have argued, even fail to separate between de re and de dicto readings (Geurts 2000a). The type of interpretation discussed in the foregoing is problematic not only for in situ accounts, but also for theories that treat specificity as a scope phenomenon. For if specificity could be dealt with in terms of scope, we should find the classical de re and de dicto readings, but not the mixed construals exemplified by (32). Specific indefinites cannot be construed in situ any more than presuppositional expressions

20

can; movement of some sort is called for. Nor is specificity a matter of scope in the grammarians’ sense: it is a pragmatic phenomenon, again like presupposition.5 4.5 Summing up It will be evident that this analysis of specificity owes a great deal to van Geenhoven’s proposal. But my account improves upon van Geenhoven’s by giving a coherent picture of the relation between specificity, on the one hand, and presupposition and definiteness, on the other, while forgoing the premise that indefinites are ambiguous between specific and nonspecific readings. Apart from providing a principled way of dealing with specificity, the present theory offers another attraction as well, in that it may shed new light on matters not directly related to specificity. It is to these matters that we now turn.

5 Second thoughts about presuppositions (and sundry other matters) Being an extension of standard DRT, the binding theory regards presuppositions as elements that would like to be bound an antecedent. This is a view that agrees with pre-theoretical intuitions about the definite article, for example, but it doesn’t seem right for some other expressions and constructions that are standardly categorized as presupposition inducers. I want to propose that at least some of these are better viewed as instances of backgrounding. 5.1 Lexical ‘presuppositions’ Intuitively speaking, the notion that presuppositions are anaphoroid elements does not seem to be quite appropriate for dealing with lexical inferences like the following, which have often been said to be presuppositional in nature (here ‘>>’ is to be read as ‘implies, intuitively speaking’): (34 ) a. Leslie is a bachelor. >> b. Leslie is a man. (35 ) a. Wilma managed to fry an egg. >> b. It was difficult for Wilma to fry an egg.

5

Many authors would balk at this and insist that presupposition is a ‘discourse-semantic’ phenomenon. Thus far

I have failed to grasp exactly what that is, but as discourse semantics seems to comprise large tracts of what I take to be pragmatics, the difference of opinion is merely verbal, I suspect.

21

(36 ) a. Fred accused Barney of nepotism. >> b. Nepotism is a bad thing. It is commonly held that (34a) presupposes (34b), and this claim seems justified by the observation that this inference tends to go through even when (34a) is embedded in nonentailing environments, such as: (37 ) Perhaps Leslie is a bachelor. A naive account of facts like this would be to suppose that the lexical content of bachelor falls into two parts: an assertional part which specifies that bachelor is truthfully predicated only of unmarried individuals, and a presuppositional part which says, among other things, that a bachelor is a man; of course, it is the second half of the content of bachelor that triggers the presupposition in (34a) and (37). There are several problems with this naive account. First, as it stands, this analysis implies that every occurrence of bachelor gives rise to the presupposition that the individual it is being applied to is a man, and therefore it predicts, for instance, that (38 ) Betty is allergic to bachelors. means something like, ‘Betty is allergic to unmarried individuals who are presupposed to be men’— which is not what we want. The solution to this problem is fairly obvious: the word bachelor should only be allowed to trigger its presupposition when it is being used predicatively. But this seems to entail that bachelor is ambiguous between a presupposing and a non-presupposing reading, which is not exactly an appealing consequence. The second problem, which is related to the first, is the following. Suppose that it is encoded in the lexicon that predicating bachelor of some individual a carries with it the presupposition that a is man. Consider now how the words bachelor and man are related to each other: the former is a hyponym of the latter, and the only distinctive feature of the word bachelor is that it applies to unmarried individuals. But at the same time that is all we are saying, as opposed to presupposing, when we call somebody a bachelor. Could this be an coincidence? I think it is pretty clear that it is not. For one thing, other hyponyms behave alike: spinster presupposes ‘female’, woodpecker presupposes ‘bird’, and so on. For another, an intuitively plausible story about this phenomenon is readily available: if a speaker wants to announce that Leslie is unmarried and has even the slightest doubt about Leslie’s sex he would say that Leslie is unmarried rather than risking (34a). I do not want to suggest that spelling out an explanation along these lines is going to be trivial, but it is obvious that if such an account could be made to work it would be much more attractive than the one we started out with, which says, in effect, that it is a lexical accident that (34a) presupposes (34b).

22

There is yet another problem with the suggestion that (predicative) bachelor presupposes ‘adult male’. It is that this presupposition, if it is one, is evidently not the kind of thing that seeks to be bound in anything like the way anaphoric elements seek to be bound. This becomes quite apparent when one considers how the binding theory would deal with (37), for example: (39 ) a. [x: Leslie(x), perhaps: [: male(x), adult(x), unmarried(x)]] b. [x: Leslie(x), male(x), adult(x), perhaps: [: unmarried(x)]] Assuming that (39a) is the semantic representation associated with (37) by the grammar, the binding theory predicts that the presupposition triggered by bachelor is accommodated in the principal DRS, because it cannot be bound and there is no reason (let us suppose) why it should be accommodated locally. This yields the right interpretation (and as a matter of fact I don’t know of any counterexamples to this analysis of bachelor), but within the framework of the binding theory this analysis causes something of an embarrassment. The presupposition supposedly triggered by bachelor can never be bound, as there is nothing to bind, so this presupposition would be one that, by its very nature, must always be accommodated, and as I have argued in my discussion of van Geenhoven’s account of specificity, that is practically a contradiction in terms. The presuppositions allegedly triggered by verbs such as manage and accuse (cf. examples (35) and (36)) are dubious, too, and partly for the same reasons. Most importantly, it just doesn’t seem to be plausible, from a pre-theoretical vantage point, that the inferences licensed by these verbs should be of an anaphoric nature, and this suspicion is strengthened by the observation that it is next to impossible to come up with examples in which these purported presuppositions must be interpreted by way of binding. My proposal is to deal with the lexical inferences in (34)–(36) in terms of backgrounding instead of presupposition. According to the theory developed in the last section, backgrounded material may be given (i.e. presupposed) but backgrounding isn’t wedded to givenness, and therefore new information may be backgrounded, too. This, it seems to me, is precisely what we witness in the cases under discussion. For example, if a speaker utters (34a), it is likely that the essential bit of information he intends to convey is that Leslie is married, not that Leslie is an adult male. Therefore, the information that Leslie is a man is backgrounded, which means, I have argued, that it will gravitate towards the principal DRS, by virtue of the Buoyancy Principle. Similarly, if someone utters (35a), he conveys (35b), but he doesn’t present this information as given (not necessarily, anyway). However, by using this particular expression, the speaker does indicate that the truth of (35b) is of less concern to him than the fact that Wilma fried an egg. Hence, even if (35b) isn’t given, we may assume that is backgrounded. The same, mutatis mutandis, for (36a). I have proposed that the lexical inferences in (34)–(36) be explained in terms of backgrounding. This is not to suggest, however, that these inferences are alike in all respects,

23

because they aren’t. Speakers’ intuitions make a fairly clear distinction between (34), on the one hand, and (35) and (36), on the other. Most speakers would say that if Leslie is a woman, (34a) is false. Whereas, if it turns out to be easy for Wilma to fry an egg, then it is not so evident what we should say about (35a). This statement would be misleading, to be sure, but many speakers would hesitate to simply reject it as false; similarly for (36a). One might say that, in contradistinction to the lexical entailment in (34), the inferences in (35) and (36) are conventional implicatures, but in view of the notorious ill-definedness of the concept of conventional implicature, that would do little more than rephrase the problem. I don’t have particularly strong opinions on how the differences between (34) and (35)– (36) can be accounted for, nor am I convinced that this issue is extremely urgent. This, however, is as it may be, because what I proposed in the foregoing doesn’t entail that such differences couldn’t exist. But these observations reinforce the suspicion voiced in section 4.1, that there may be various ways of backgrounding, which may not all be equivalent. 5.2 Presupposition vs. background Over the past few decades, but especially during the presupposition craze of the seventies, the label ‘presuppositional’ has been applied to such a bewildering variety of phenomena that the very notion of presupposition has become suspect, as the following passage from Neale (1990: 54) illustrates: A great range of disparate and unrelated phenomena has been dubbed ‘presuppositional’ over the years, but [...] it seems highly implausible that any theoretically important notion will do justice to the full range of data that semanticists professing an interest in ‘presupposition’ seek to explain. Needless to say, I am not entirely convinced that the second half of this claim is justified, but the first half certainly is. All too often, the concept of presupposition has been used, or rather abused, without even the shadow of justification. I have argued elsewhere that this abuse was caused at least in part because the diagnostic tests for presuppositionhood were (and still are) applied too carelessly, if they were applied at all (see Geurts 1999). But in the light of the foregoing discussion I want to suggest that there may have been another factor as well, which is that the standard tests don’t allow us to make a clear distinction between presuppositional and backgrounding effects, and that at least some of the phenomena that have been categorized, to greater or lesser acclaim, as ‘presuppositional’ are better seen in terms of backgrounding. The lexical inferences discussed previously are relatively clear instances of this category, and further possible candidates for relocation will be discussed below. But first I want to raise the question how we are going to distinguish between genuine presuppositions and instances of backgrounding.

24

This is not a trivial question because, as I hinted already, the standard litmus tests for presuppositionhood fail to distinguish between presupposition and backgrounding, as the following observations illustrate: (40 ) a. If Germany becomes a monarchy again, the king of France will get nervous >> There is a king of France. b. If there is a king of France, the king of France will get nervous >> / There is a king of France. (41 ) a. If Leslie is rich, he is a bachelor >> Leslie is a man. b. If Leslie is a man, he is a bachelor >> / Leslie is a man. (42 ) a. If the king of France gets nervous, his ministers get nervous, too. >> There is a king of France. b. If the king of France gets nervous, then France must be a monarchy >> / There is a king of France. (43 ) a. If Leslie is a bachelor, he is rich >> Leslie is a man. b. If Leslie is a bachelor, he is a man >> / Leslie is a man. These observations suggest that there are no differences between the presupposition triggered by the definite NP ‘the king of France’, as in (40) and (42), and the lexical inference licensed by the noun bachelor, as in (41) and (43), and the parallels extend to all sorts of embedding contexts. Nevertheless, I have argued, there are good reasons for believing that lexical inferences aren’t of a presuppositional nature. But none of these reasons provides us with a general criterion for discriminating between presupposition and backgrounding. According to the binding theory, presupposed information is presented as given, in the same sense that the antecedent of an anaphoric expression is given, and the theory’s treatment of presupposition is a generalization of DRT’s treatment of anaphora, which is based on the widely held view that an anaphoric expression serves to retrieve an element from the common ground. That is to say, the speaker employs an anaphor not merely to signal that a discourse entity x is given, but also as an instruction to the hearer that he should identify and recover the intended x, so that new information will have the right connections. In other words, the hearer is expected to ask himself which entity the speaker has in mind. I want to suggest that we can turn this observation into a useful test for distinguishing between real presuppositions and merely backgrounded information. The test goes as follows: If x is a genuine presupposition, then it should make sense to ask ‘Which x do you mean?’ when the speaker has just uttered a sentence implying the existence of some x. This admittedly informal criterion indicates that, for example, the following are genuine presupposition inducers:

25

• Pronouns: (44 ) A: He is insane. B: Who is insane? • Definite NP’s: (45 ) A: The banana has been stolen. B: Which banana has been stolen? • Quantifier domains: (46 ) A: Every girl has sent me a postcard. B: Which girls have sent you a postcard? • Focus particles: (47 ) A: Professor Babel has read my paper, too. B: Who else has read your paper? On the other hand, there are various alleged presupposition inducers that fail the wh-test. The lexical inferences discussed in the previous section are a case in point, as are factive verbs and transition verbs, for example, which are standardly listed among the presupposition-inducing expressions: • Factives: (48 ) a. Barney is proud that his daughter is an anarcho-syndicalist. >> b. Barney’s daughter is an anarcho-syndicalist. • Transition verbs: (49 ) a. Betty has started taking saxophone lessons [at time t]. >> b. Betty wasn’t taking saxophone lessons [before t]. The inferences in (48) and (49) originate with the factive be proud and the transition verb start, respectively, and they both exhibit the projection behaviour that is characteristic of presuppositions. But they also fail the wh-test. In the first case it would make no sense to ask which state or fact (or whatever) involving his daughter Barney is proud of, and in the second case no hearer would ever wonder which instance of Betty-not-taking-saxophone-lessons

26

ended at time t. Hence, if the wh-test is to be trusted, the inferences exemplified by (48) and (49) aren’t genuine presuppositions, and therefore they must be explained in terms of backgrounding. Zeevat (1992) has proposed a classification of presupposition-inducing expressions which resembles my somewhat tentative distinction between genuine presupposition inducers and expressions licensing inferences that are best understood in terms of backgrounding. Zeevat’s ‘resolution triggers’ correspond to what I call ‘presuppositions’ simpliciter; his ‘lexical triggers’, to what I prefer to treat as backgrounding expressions.6 It would take me too far afield to discuss the theory Zeevat erects on his classification, but I would like to briefly comment on one of his empirical claims, which, if correct, might be put to use for discriminating between presupposition inducers and backgrounding expressions. Zeevat views lexical triggers as ‘applicability conditions’ which must be satisfied locally, i.e. in situ; and this constraint does not hold, according to Zeevat, for resolution triggers. It follows from this that resolution triggers can, and lexical triggers cannot, get de re construals. The following example illustrates both predictions: (50 ) Betty believes that the superintendent is a bachelor. If this statement is true, Betty can hardly fail to believe that the superintendent is a man (which is the lexical inference triggered by bachelor), but it may well be that she is not aware that the person in question is a superintendent (which is part of the presupposition triggered by the definite NP). Unfortunately however, for Zeevat as well as myself, this distinction is not as neat as it initially appears to be. Suppose that all Betty knows about the superintendent is that he or she is not married. Would (50) be true or false, under these circumstances? Speaking for myself, I believe I might accept the statement as true, but even if other speakers should disagree, they would still have to concede, I think, that the matter is not as clear-cut as it seemed to be at first. When we turn away from the standard bachelor-type cases, it becomes even clearer that Zeevat’s observation is hard to maintain. Suppose Fred tells his friend Barney: ‘Wilma fried an egg this morning.’ Whereupon Barney reports to his wife: (51 ) Fred believes that Wilma managed to fry an egg. Tendentious though it may be, this statement is clearly correct, and it need not imply that Fred believes that it is (or was) difficult for Wilma to fry an egg. Therefore, if Zeevat’s diagnostic applied across the board, this inference could not be a lexical presupposition (in

6

The correspondences are not quite perfect, though. More recently, Zeevat (to appear) has presented a new and

finer-grained classification of presupposition triggers, which is rather less similar to what is proposed here than his 1992 version.

27

Zeevat’s terminology) or backgrounded information (in mine). I don’t know how Zeevat would want to deal with this inference, but since I want to treat it as an instance of backgrounding, I cannot employ attitude contexts for distinguishing between presuppositions and backgrounded information. 5.3 Factives Factive verbs are standardly regarded as presupposition-inducing expressions, although there is a well-known problem with this view. It is that some factive verbs, at least, do not always seem to trigger the presupposition that their complement is true: (52 ) a. If Barney should discover that Miss Chambley is rich, he’ll propose to her. b. If I should discover that Miss Chambley is rich, I’ll propose to her. Both (52a) and (52b) can be consistently uttered by a speaker who doesn’t want to commit himself as to whether Miss Chambley is rich, but unlike (52b), (52a) appears to have a further reading, as well, implying that Miss Chambley is rich. In view of observations such as these it has been suggested that discover belongs to a special class of ‘semi-factive’ verbs, which are ambiguous between a presupposing and a non-presupposing reading. This unattractive assumption can be avoided if we approach the matter in somewhat different terms. If the complement of a factive verb can be either backgrounded or not, the Buoyancy Principle predicts that something very much like presupposition projection will occur in the former case but not in the latter. This view is an attractive one, I believe, because it seems to correlate with our intuitions about foreground vs. background in factive constructions. For example, a speaker who utters (53a) may be interested primarily in the fact that Barney knew (53b), or in the fact that (53b) is true. In the former case, the information in (53b) is backgrounded; in the latter, it is foregrounded. (53 ) a. Barney knows that his daughter is an anarcho-syndicalist. >> b. Barney’s daughter is an anarcho-syndicalist. Now if the same options are available in (52), we predict that backgrounding the proposition that Miss Chambley is rich will imply that Miss Chambley is rich, whereas this inference will not go through if the factive complement is foregrounded. These predictions appear to be correct. 5.4 Concluding remarks In the preceding pages I have argued that a number of expressions that are standardly categorized as presupposition inducers are better viewed as backgrounding devices. I suspect,

28

furthermore, that this viewpoint may be of more general use, and that it may help to account for phenomena which have not as yet received a satisfactory treatment. Let me mention just two, rather disparate, examples: • Non-restrictive relative clauses: (54 ) a. Fred suspected that Betty, who had been avoiding him of late, had discovered about his collection of Neil Sedaka albums. >> b. Betty had been avoiding Fred of late. • Felicity conditions on speech acts: (55 ) a. Where is my bicycle? >> b. The speaker doesn’t know where his bicycle is. Although it has occasionally been suggested that these inferences are of a presuppositional nature, this position has not gained much support in the literature (exceptions are Fillmore 1969 and Keenan 1971). Still, both types of inference seem to exhibit the ‘wide scope’ tendency that is the hallmark of presuppositions. This is harder to demonstrate for felicity conditions on speech acts than for non-restrictive relatives, because non-declaratives dislike being embedded under operators of any kind. But at least we have conditional speech acts: (56 ) a. If my pogo stick is in the attic, where is my bicycle? >> b. The speaker doesn’t know where his bicycle is. That non-restrictive relative clauses behave similarly is easier to show, for instance, by embedding (54) under a weak modal operator, such as perhaps. Nevertheless, it seems to me that the majority view is correct, and that the inferences exemplified by (54) and (55) shouldn’t be granted the status of presuppositions. In particular, the preferred interpretation of non-restrictive relatives is plausibly explained in terms of backgrounding: non-restrictive relatives are parenthetical remarks, which are backgrounded if anything is. So the Buoyancy Principle surely applies to non-restrictive relatives, and I conjecture that it applies to felicity conditions on speech acts, too.

References Abbott, B. 2000: Presuppositions as nonassertions. Journal of Pragmatics 32: 1419-1437. Abusch, D. 1994: The scope of indefinites. Natural Language Semantics 2: 83-135. Cresti, D. 1995: Indefinite topics. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, Mass. Farkas, D.F. 1981: Quantifier scope and syntactic islands. Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society. Pp. 59-66.

29

Fillmore, C.J. 1969: Types of lexical information. In: F. Kiefer (ed.), Studies in Syntax and Semantics. Reidel, Dordrecht. Pp. 109-137. Fodor, J.D. and I.A. Sag 1982: Referential and quantificational indefinites. Linguistics and Philosophy 5: 355398. Foley, W.A. 1994: Information structure. In: R.E. Asher (ed.), The Encyclopedia of Language and Linguistics, Vol. 3. Pergamon Press, Oxford. Pp. 1678-1685. Foley, W.A. and R.D. van Valin 1985: Information packaging in the clause. In: T. Shopen (ed.), Language Typology and Syntactic Description. Volume I: Clause Structure. Cambridge University Press. Pp. 282-364. Geurts, B. 1998a: The mechanisms of denial. Language 74: 274-307. Geurts, B. 1998b: Presuppositions and anaphors in attitude contexts. Linguistics and Philosophy 21: 545-601. Geurts, B. 1999: Presuppositions and Pronouns. Elsevier, Oxford. Geurts, B. 2000a: Indefinites and choice functions. Linguistic Inquiry 31: 731-738. Geurts, B. 2000b: Buoyancy and strength. Journal of Semantics 17: 315-333. Geurts, B. and R. van der Sandt, 1999: Domain restriction. In: P. Bosch and R. van der Sandt (eds.), Focus: Linguistic, Cognitive, and Computational Perspectives. Cambridge University Press. Pp. 268-292. Givón, T. 1978: Definiteness and referentiality. In: J.H. Greenberg, C.A. Ferguson, and E.A. Moravcsik (eds.), Universals of Human Language, Vol. 4: Syntax. Stanford University Press. Pp. 291-330. Givón, T. 1984: Syntax: A Functional-Typological Introduction. Volume 1. Benjamins, Amsterdam. Haspelmath, M. 1997: Indefinite Pronouns. Oxford University Press. Heim, I. 1983: On the projection problem for presuppositions. Proceedings of the West Coast Conference on Formal Linguistics 2: 114-126. Heim, I. 1991: Artikel und Definitheit. In: A. von Stechow and D. Wunderlich (eds.), Semantics: An International Handbook of Contemporary Research. Walter de Gruyter, Berlin. Pp. 487-535. Heim, I. 1992: Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9: 183-221. Horn, L.R. 2000: Assertoric inertia. Paper presented at the LSA meeting, Chicago. Kamp, H. 1981: A theory of truth and semantic representation. In: J.A.G. Groenendijk, T.M.V. Janssen, and M.B.J. Stokhof (eds.), Formal Methods in the Study of Language. Mathematical Centre Tracts 135, Amsterdam. Pp. 277-322. Kasher, A. and D.M. Gabbay 1976: On the semantics and pragmatics of specific and non-specific indefinite expressions. Theoretical Linguistics 3: 145-190. Keenan, E.L. 1971: Two kinds of presupposition in natural language. In: C.J. Fillmore and D.T. Langendoen (eds.), Studies in Linguistic Semantics. Holt, New York. Pp. 45-54. Kratzer, A. 1995: Scope or pseudoscope? Are there wide-scope indefinites? In: S. Rothstein (ed.), Events and Grammar. Kluwer, Dordrecht. Pp. 163-196. Kuno, S. 1987: Functional Syntax: Anaphora, Discourse and Empathy. University of Chicago Press. Kuroda, S.-Y. 1981: Indexed predicate logic. Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society. Pp. 156-163. Krifka, M., F.J. Pelletier, G.N. Carlson, A. ter Meulen, G. Link, and G. Chierchia 1995: Genericity: an introduction. In: G.N. Carlson and F.J. Pelletier (eds.), The Generic Book. University of Chicago Press. Pp. 1-124. Ladusaw, W. 1982: Semantic constraints on the English partitive construction. Proceedings of the 1st West Coast Conference on Formal Linguistics, 231-242. Lyons, C. 1999: Definiteness. Cambridge University Press. Manga, L. 1996: An Explanation for Ergative versus Accusative Languages: An Examination of Inuktitut. Doctoral thesis, University of Ottawa. Matthewson, L. 1999: On the interpretation of wide-scope indefinites. Natural Language Semantics 7: 79-134. Neale, S. 1990: Descriptions. MIT Press, Cambridge, Mass. Prince, E. 1981: On the inferencing of indefinite-this NPs. In: A.K. Joshi, B.L. Webber, and I.A. Sag (eds.), Elements of Discourse Understanding. Cambridge University Press. Pp. 231-250. Reinhart, T. 1997: Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335-397. Stalnaker, R.C. 1984: Inquiry. MIT Press, Cambridge, Mass. Van der Sandt, R. 1992: Presupposition projection as anaphora resolution. Journal of Semantics 9: 333-377.

30

Van der Sandt, R. and B. Geurts 2001: Too. Proceedings of the 13th Amsterdam Colloquium. University of Amsterdam. Van Geenhoven, V. 1998: Semantic Incorporation and Indefinite Descriptions. CSLI Publications, Stanford. Van Rooy, R. 1999: The specificity of indefinites. Ms, University of Amsterdam. Winter, Y. 1997: Choice functions and the scopal semantics of indefinites. Linguistics and Philosophy 20: 399467. Yeom, J.-I. 1998: A Presuppositional Analysis of Specific Indefinites: Common Grounds as Structured Information States. Garland, New York. Zeevat, H. 1992: Presupposition and accommodation in update semantics. Journal of Semantics 9: 379-412. Zeevat, H. to appear: Explaining presupposition triggers. In: K. van Deemter and R. Kibble (eds.), Information Sharing. CSLI, Stanford.

31